19.07.2014 Views

Contents - Student subdomain for University of Bath

Contents - Student subdomain for University of Bath

Contents - Student subdomain for University of Bath

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

2.1. WHAT ARE POLYNOMIALS? 37<br />

While characterisations 3 and 2 <strong>of</strong> a Gröbner base (theorem 13) can make sense<br />

in either view, characterisations 4 and 1 (the only effective one) only make sense<br />

in a distributed view. Conversely, while the abstract definitions <strong>of</strong> factorisation<br />

and greatest common divisors (definition 25) make sense whatever the view, the<br />

only known algorithms <strong>for</strong> computing them (algorithm 1 or the advanced ones<br />

in chapter 4) are inherently recursive 15 .<br />

2.1.5 Other representations<br />

Sparse representations take up little space if the polynomial is sparse. But<br />

shifting the origin from x = 0 to x = 1, say, will destroy this sparsity, as<br />

might many other operations. The following example, adapted from [CGH + 03],<br />

illustrates this. Let Φ(Y, T ) be<br />

∃X 1 . . . ∃X n (X 1 = T + 1) ∧ (X 2 = X 2 1 ) ∧ · · · ∧ (X n = X 2 n−1) ∧ (Y = X 2 n). (2.4)<br />

The technology described in section 3.5.2 will convert this to a polynomial equation<br />

Ψ(Y, T ) : Y = (1 + T ) 2n . (2.5)<br />

Dense or sparse representations have problems with this, in the sense that expression<br />

(2.4) has length O(n), but expression (2.5) has length O(2 n ) or more.<br />

A factored representation could handle the right-hand side, assuming that we<br />

are not representing the equations as polynomial = 0. But changing the last<br />

conjunct <strong>of</strong> Φ to (Y = (X n + 1) 2 ) changes Ψ to<br />

Y =<br />

(<br />

1 + (1 + T ) 2n−1) 2<br />

, (2.6)<br />

whose factored representation now has length O(2 n ).<br />

Factored representations display a certain amount <strong>of</strong> internal structure, but<br />

at the cost <strong>of</strong> an expensive, and possibly data-expanding, process <strong>of</strong> addition.<br />

Are there representations which do not have these ‘defects’? Yes, though they<br />

may have other ‘defects’.<br />

Expression tree This representation “solves” the cost <strong>of</strong> addition in the factored<br />

representation, by storing addition as such, just as the factored<br />

representation stored multiplication as such. Hence ( (x + 1) 3 − 1 ) 2<br />

would<br />

be legal, and represented as such. Equation (2.6) would also be stored<br />

compactly provided exponentiation is stored as such, e.g. Z 2 requiring<br />

one copy <strong>of</strong> Z, rather than two as in Z · Z. This system is not canonical,<br />

or even normal: consider (x + 1)(x − 1) − (x 2 − 1). This would be<br />

described by Moses [Mos71] as a “liberal” system, and generally comes<br />

with some kind <strong>of</strong> expand command to convert to a canonical representation.<br />

Assuming now that the leaf nodes are constants and variables,<br />

15 At least <strong>for</strong> commutative polynomials. Factorisation <strong>of</strong> non-commutative polynomials is<br />

best done in a distributed <strong>for</strong>m.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!